Latest -current complete lockup (tcp changes?)

[Robert Watson]

On Tue, 27 Mar 2007, Andre Oppermann wrote:

> Andrey Chernov wrote:
>> On Tue, Mar 27, 2007 at 09:28:10AM +0400, Andrey Chernov wrote:
>>> On Tue, Mar 27, 2007 at 10:11:01PM +0000, Garrett Cooper wrote:
>>>>> The problem is deeper than that (((
>>>>> I still got the same lockup, just with more net activity.
>>>>> I even try to completely disable sack, with the same result, so probem 
>>>>> is somewhere else. Last working kernel still from Mar 22.
>>>> I'll give a CVSup / upgrade a try and see what happens.
>>> Additional non-default details from my machine which may (or may not) 
>>> affect the thing:
>> 
>> Yet one detail about lockup: external pings to the machine works in the 
>> lockup situation, but no any TCP services is available.
>
> I can only think of a TCP_INFO_LOCK() leak here blocking any further 
> progress.  When I'm back in the office in about an hour I'll prepare a test 
> patch.

If the tcbinfo lock were being leaked, then the netisr or ithread would remain 
permanently wedged preventing any IP services from working, and rapidly hosing 
the machine as the ithread wouldn't be doing its thing.  I think another sort 
of problem is more likely.

Robert N M Watson
Computer Laboratory
University of Cambridge